Anesthesia nerve precise positioning method and system for ultrasonic image auxiliary processing

Abstract

The invention discloses an anesthesia nerve precise positioning method and system based on ultrasonic image auxiliary processing, and the method comprises the following steps: obtaining a real-time dynamic ultrasonic image sequence of a target region through an ultrasonic image data collection device, carrying out the multi-modal data fusion processing, and generating an enhanced ultrasonic image data set; performing three-dimensional semantic segmentation on the enhanced ultrasonic image data set based on a preset anesthesia nerve region segmentation model, and extracting morphological characteristics and spatial distribution information of a target nerve tissue; generating a three-dimensional navigation path grid based on the extracted morphological characteristics and spatial distribution information, and constructing a real-time visual guide interface; and dynamically optimizing a navigation path and a risk early warning strategy based on real-time operation feedback. The method has the following advantages and effects: a precise positioning method capable of fusing multi-scale neural features and dynamically optimizing a navigation path is provided, so that the key technical bottleneck of nerve ending visualization and real-time obstacle avoidance is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method and system for accurately locating anesthetic nerves using ultrasound image-assisted processing. Background Art

[0002] In ultrasound-guided anesthesia nerve localization, how to accurately distinguish between the main nerve trunk and its peripheral branches and plan a safe puncture path in real time is a core issue that needs to be addressed. Traditional ultrasound-guided methods rely mainly on the physician's experience to manually identify nerve structures. However, because the peripheral branches of nerves appear as weak signals with low contrast and high noise in ultrasound images, their morphological characteristics are easily affected by surrounding blood vessels and soft tissue artifacts, resulting in the following key defects in intraoperative puncture path planning: High missed detection rate of peripheral branches: Existing segmentation algorithms are not sensitive enough to nerve endings and cannot accurately extract their spatial distribution, causing the navigation path to stray into high-risk areas; Static path defect: The traditional navigation path is statically generated based on preoperative images and cannot be dynamically adjusted according to intraoperative instrument movement and tissue deformation, causing the actual puncture trajectory to deviate from the preset safety boundary.

[0003] The above problems directly lead to an increased risk of intraoperative nerve injury, especially in high-density nerve plexus areas, where slight operational deviations can lead to irreversible complications. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for accurately locating anesthetic nerves using ultrasound imaging as an aid to processing, so as to solve the problems raised in the background technology.

[0005] The above technical objectives of the present invention are achieved through the following technical solutions: The present invention provides a method for accurately locating anesthesia nerves using ultrasound imaging-assisted processing, comprising the following steps: S100, acquiring a real-time dynamic ultrasound image sequence of a target area through an ultrasound image data acquisition device, and performing multimodal data fusion processing on the real-time dynamic ultrasound image sequence to generate an enhanced ultrasound image dataset; wherein the multimodal data fusion processing includes spatial resolution enhancement, noise suppression, and tissue contrast optimization; S200, performing three-dimensional semantic segmentation on the enhanced ultrasound image dataset based on a preset anesthesia nerve region segmentation model to extract morphological features and spatial distribution information of target nerve tissue; wherein the anesthesia nerve region segmentation model adopts a cascade convolutional neural network structure; S300, generating a three-dimensional navigation path grid based on the extracted morphological features and spatial distribution information, and constructing a real-time visual guidance interface; wherein the three-dimensional navigation path grid includes spatial coordinate information of the recommended puncture path of the nerve trunk, the prohibited intrusion area of ​​the peripheral branch, and the safe operation boundary; S400, dynamically optimize navigation paths and risk warning strategies based on real-time operation feedback.

[0006] By adopting the above technical solutions, through multimodal data fusion processing, integrating spatial resolution enhancement, noise suppression and tissue contrast optimization technologies, the visualization effect of neural structures in ultrasound images can be significantly improved; the enhanced image data set can simultaneously retain the macroscopic morphology of the nerve trunk and the microscopic details of the terminal branches, providing high-quality input for subsequent precise segmentation and navigation path planning; multimodal fusion technology effectively overcomes the limitations of a single processing method, noise suppression technology distinguishes signal and noise components through frequency domain analysis, and tissue contrast optimization technology strengthens the boundary characteristics of the target structure through regional processing; a cascaded convolutional neural network structure is used to perform three-dimensional semantic segmentation on the enhanced image, effectively distinguishing the grayscale differences between neural tissue and surrounding blood vessels and soft tissue, reducing the risk of missegmentation; the segmentation results contain complete neural topology information, which directly supports the precise labeling of avoidance areas in subsequent path planning. ; The design of the cascade network uses a phased processing strategy to first locate the nerve trunk area and then refine the terminal branch structure, avoiding feature confusion caused by one-time segmentation of complex targets; three-dimensional semantic segmentation not only provides voxel-level classification results, but also retains the spatial continuity of neural tissue, laying the foundation for morphological feature extraction; based on the three-dimensional navigation path grid generated by neural morphological characteristics and spatial distribution, the recommended puncture path, prohibited intrusion area and safe operation boundary of the nerve trunk are annotated in real time to reduce the cognitive load of the operator during operation; combined with the dynamic optimization mechanism of real-time feedback, it ensures that the navigation path is adaptively adjusted with the movement of the instrument and tissue deformation, thereby improving the safety of puncture; the core of dynamic optimization is to collect the instrument motion trajectory in real time and perform matching analysis with the preset path. When an offset or unstable operation is detected, the path correction and risk warning are immediately triggered, thereby improving the real-time and reliability of intraoperative navigation.

[0007] Further configuration is that the S100 specifically includes the following sub-steps: S110, performing dynamic noise suppression processing on the real-time dynamic ultrasound image sequence, constructing an adaptive filtering template based on regional sensitivity by analyzing the spectral distribution characteristics of the noise in the image; wherein the adaptive filtering template dynamically adjusts the filtering strength according to the gradient amplitude of the local pixel, reducing the filtering strength in nerve-dense areas to protect microstructural details, and increasing the filtering strength in homogeneous tissue areas to eliminate noise interference; S120, performing multi-scale spatial resolution optimization on the noise-suppressed ultrasound image to fuse local texture features and global structural information of the image; wherein the multi-scale optimization includes: extracting the spatial distribution pattern of the nerve trunk at a coarse scale level to generate a neural pathway skeleton at a low resolution; combining the phase information of the ultrasound echo signal at a fine scale level to restore the morphological characteristics of the nerve terminal branches; and fusing the coarse-scale and fine-scale reconstruction results through an iterative optimization algorithm to generate a high-fidelity neural image; S130, performing region-by-region contrast enhancement processing on the high-fidelity neural image, including: dividing the high-fidelity neural image into a neural core area, a vascular distribution area, and a background soft tissue area, and defining the grayscale dynamic range of each area; A local histogram stretching algorithm is used in the nerve core area to enhance the boundary contrast between the perineurium and myelin sheath; The overlapping artifacts between the vascular wall and the epineurium are suppressed by a nonlinear mapping function in the vascular distribution area; Apply global contrast compression algorithm in the background soft tissue area to reduce the visual interference of irrelevant tissues; The enhancement results of each region are spatially fused to generate the enhanced ultrasound image dataset.

[0008] By adopting the above technical solution, a regional sensitivity filter template is constructed by analyzing the noise spectrum characteristics, and the filter strength is dynamically adjusted to suppress high-frequency speckle noise and low-frequency motion artifacts while preserving the continuity of the nerve edge. This method overcomes the microstructural blurring problem caused by traditional global filtering and improves the clarity of the nerve boundary. The spatial distribution pattern of the nerve trunk is extracted at the coarse scale level to generate a low-resolution skeleton. At the fine scale level, the morphology of the terminal branches is restored by combining the ultrasound echo phase information, and the multi-scale features are integrated through an iterative optimization algorithm to solve the problem of peripheral signal loss caused by the ultrasound diffraction effect. The nerve core area, vascular distribution area and background soft tissue area are divided, and the contrast of each area is optimized using local histogram stretching, nonlinear mapping and global compression algorithms respectively. The boundary contrast enhancement of the nerve core area improves the surgeon's ability to identify key structures, the suppression of overlapping artifacts in the vascular area reduces the risk of misjudgment, and the reduction of visual interference in the background area further focuses the operating field of view.

[0009] Further configuration is that the S200 specifically includes the following sub-steps: S210, constructing a cascade convolutional neural network model, wherein the cascade convolutional neural network model includes a first-level coarse segmentation network, a second-level fine segmentation network, and a cross-level attention fusion module; S220, performing three-dimensional voxel processing on the enhanced ultrasound image data set to generate standardized voxel data blocks; S230, inputting the voxel data block into the cascade convolutional neural network model, performing three-dimensional semantic segmentation, generating a first-level feature map and a second-level feature map, and performing weighted fusion; S240, generating a three-dimensional neural region mask based on the weighted fusion of the first-level feature map and the second-level feature map, and optimizing the segmentation result through morphological post-processing; S250, extracting the morphological characteristics and spatial distribution information of the target neural tissue from the three-dimensional neural region mask; specifically comprising the following sub-steps: Based on the spatial connectivity of voxels in the three-dimensional neural region mask, the centerline path of the neural trunk and its curvature distribution are calculated to generate a neural direction vector field as a morphological feature; Extracting the topological structure of the candidate nerve terminal branches using a three-dimensional skeletonization algorithm to generate three-dimensional skeletonized branches as spatial distribution information, and quantifying branch lengths, bifurcation angles, and diameter change rates; Combined with the spatial coordinate system defined in the enhanced ultrasound image dataset, the nerve direction vector field and the three-dimensional skeletonized branches are mapped to the corresponding anatomical positions, a density distribution heat map of the neural tissue is constructed, and the geometric center coordinates of the high-density neural plexus area are marked; wherein the high-density neural plexus area is defined by a voxel cluster whose probability density value exceeds a preset threshold in the heat map.

[0010] By adopting the above technical solutions, the first-level coarse segmentation network quickly locates the neural trunk area, the second-level fine segmentation network expands the receptive field through void convolution to cover the complex branch topology, and the cross-level attention fusion module weightedly integrates multi-scale features to highlight the semantic information of the neural core area; this design significantly improves the detection rate of terminal branches while ensuring the segmentation speed; the coarse segmentation network adopts a lightweight structure to reduce computational overhead, and the fine segmentation network extracts fine-grained features through a deep architecture; the attention mechanism dynamically assigns weights according to the spatial correlation of the feature map, making the feature expression of important areas more significant and reducing the impact of background interference on the segmentation results; the precise recording of spatial coordinate information ensures that the segmentation results are highly consistent with the actual anatomical structure, which provides space for subsequent navigation paths. The voxelization process converts the two-dimensional slice sequence into three-dimensional data, preserving the spatial continuity information. The bicubic interpolation algorithm compensates for the inter-layer misalignment caused by probe movement or tissue deformation by fitting the grayscale distribution of adjacent slices, ensuring the realization of isotropic resolution in the voxel grid. The three-dimensional connected domain analysis is used to remove isolated noise points and retain continuous neural pathways. The candidate terminal branches are screened based on the preset diameter threshold to eliminate the false structures formed by artifact interference. The post-processing step effectively improves the reliability of the segmentation mask and reduces the risk of misjudgment in navigation path planning. The three-dimensional connected domain analysis identifies independent areas based on the spatial adjacency of voxels, and the volume threshold filtering removes small noise. The morphological opening operation smoothes the segmentation boundary and eliminates jagged artifacts.

[0011] Further configuration is that the S220 specifically includes: The axial slice sequence is divided along the ultrasound probe scanning direction, and an isotropic voxel grid is constructed based on spatial continuity constraints between adjacent slices. The grayscale values ​​within each voxel grid are optimized using bicubic interpolation to eliminate inter-slice resolution differences caused by probe motion. The optimized voxel grid is integrated through three-dimensional spatial resampling to generate standardized voxel data blocks, where each voxel contains spatial coordinate information and corresponding grayscale intensity value. The S230 specifically includes: The first-level coarse segmentation network extracts global contextual features of each axial slice to generate a first-level feature map containing the neural trunk area; the second-level fine segmentation network refines the first-level feature map layer by layer, uses dilated convolution to expand the receptive field to cover the topological structure of the neural branches, retains low-level detail information through residual connection, and outputs a second-level feature map; in the cross-level attention fusion module, the channel correlation matrix of the first-level feature map and the second-level feature map is calculated to generate a spatial attention weight map; based on the spatial attention weight map, the first-level feature map and the second-level feature map are weightedly fused to highlight the semantic information of the neural core area; The morphological post-processing in S240 includes: The first-level feature map and the second-level feature map after weighted fusion are threshold segmented to generate a three-dimensional neural region mask; isolated noise points in the three-dimensional neural region mask are removed, and continuous neural paths are retained; then, ultra-fine branches of the three-dimensional neural region mask are screened based on a preset nerve diameter threshold, candidate nerve terminal branches are retained, and false neural structures formed by artifact interference are eliminated.

[0012] By adopting the above technical solutions, a voxel grid is constructed based on the spatial continuity constraints of adjacent slices to solve the problem of inter-layer misalignment caused by probe movement; bicubic interpolation optimizes the grayscale distribution, eliminates the impact of resolution differences on segmentation accuracy, and ensures the spatial consistency of three-dimensional voxel data; the sequential arrangement of axial slices and spatial continuity analysis can detect and correct inter-layer displacement, and the interpolation algorithm restores the anatomical structure continuity lost due to probe jitter through local grayscale gradient fitting; the fine segmentation network uses void convolution layers with increasing expansion rates to cover the long-range dependencies of neural branches, and residual connections retain low-level detail information. This design avoids detail loss while expanding the receptive field, improving the completeness and accuracy of terminal branch segmentation; it generates a spatial attention weight map by calculating the channel correlation matrix, and weightedly fuses the feature maps of the coarse and fine segmentation networks; this mechanism significantly increases the fusion weight of the neural core area and reduces the weight of the background area, further enhancing the boundary clarity of the segmentation mask; the attention weight map is dynamically generated according to the channel response of the feature map, with high-response areas corresponding to key parts of the neural tissue and low-response areas corresponding to background or noise; after weighted fusion, the feature expression of the neural area is enhanced and the features of irrelevant areas are suppressed, thereby improving the semantic consistency of the segmentation results.

[0013] Further configuration is that the S300 specifically includes the following sub-steps: S310: Generate a navigation baseline for the nerve trunk and an avoidance path for the distal branches using a path optimization algorithm based on the curvature distribution of the nerve trajectory vector field and the topological structure of the three-dimensional skeletonized branches; wherein the path optimization algorithm includes: Equally spaced sampling points are set along the centerline path of the nerve trunk, and the curvature radius of each sampling point is calculated. If the curvature radius is less than a preset safety threshold, the path curvature is smoothed using a Bezier curve interpolation algorithm to avoid sudden changes in the navigation path. Constructing a dynamic avoidance buffer zone around the distal branch according to the bifurcation angle and diameter change rate of the three-dimensional skeletonized branch, wherein the radius of the buffer zone is positively correlated with the diameter of the distal branch; S320: Mapping the navigation baseline and the avoidance path to the spatial coordinate system of the anatomical position to generate a three-dimensional navigation path grid; wherein the three-dimensional navigation path grid includes spatial coordinate information of the recommended puncture path of the nerve trunk, the prohibited intrusion area of ​​the peripheral branch, and the safe operation boundary; S330: Based on the geometric center coordinates of the high-density nerve plexus region in the density distribution heat map, marking the high-risk operation area in the three-dimensional navigation path grid, and generating a risk level label through probability density gradient calculation; wherein the high-risk operation area is defined as a voxel cluster whose probability density value in the spatial coordinate system exceeds a preset threshold; S340: Dynamically spatially register the three-dimensional navigation path grid with the enhanced ultrasound image data set to generate a real-time visual guidance interface.

[0014] By adopting the above technical solution, Bezier curves are inserted into areas of sudden curvature along the nerve trunk path to generate a smooth and continuous navigation baseline, thus avoiding the risk of nerve damage caused by sudden changes in the puncture needle path. The radius of the dynamic avoidance buffer zone is positively correlated with the diameter of the distal branch, ensuring adaptive adjustment of the avoidance distance for branches of varying thicknesses. High-density nerve plexus areas are annotated based on the density distribution thermogram, and risk level labels are generated based on the probability gradient. This allows the surgeon to intuitively identify high-risk operation areas and adjust the puncture strategy in real time. The bounding box size of the safe operation boundary is dynamically generated based on the nerve trunk diameter and heat map density, covering potential risk areas. Dynamic spatial registration is used to project the three-dimensional navigation mesh onto the ultrasound image plane, and cross-sectional views are superimposed to display the recommended path, restricted areas, and safe boundaries, with real-time updates of the distance between the instrument tip and the boundary. Multi-view collaborative guidance significantly reduces the surgeon's spatial cognitive errors.

[0015] It is further configured that, in S320, the three-dimensional navigation path grid includes the spatial coordinate information of the recommended puncture path of the nerve trunk, the prohibited invasion area of ​​the peripheral branch, and the safe operation boundary, specifically: The recommended puncture path of the nerve trunk is generated by curvature smoothing the center line path of the nerve trend vector field and is represented by a sequence of three-dimensional coordinate points in the spatial coordinate system; The prohibited intrusion area of ​​the terminal branch is defined as a cylindrical area with the center line of the terminal branch as the axis and the radius of the buffer zone as the boundary according to the radius of the dynamic avoidance buffer zone and the spatial position of the terminal branch, and is marked by a voxel mask in the spatial coordinate system; The safe operation boundary is generated based on the minimum safe distance threshold between the nerve trunk and the terminal branches, and is marked in the form of a bounding box in the spatial coordinate system. The size of the bounding box is determined by the average diameter of the nerve trunk and the density distribution heat map of the high-density nerve plexus area.

[0016] By adopting the above technical solution, a cylindrical prohibited intrusion zone is constructed with the center line of the distal branch as the axis, and full coverage of the three-dimensional space is ensured through voxel mask marking; the radius of the cylindrical area is adaptively adjusted according to the diameter of the distal branch, with coarse branches corresponding to larger buffer zones and thin branches corresponding to smaller buffer zones, balancing safety and operating space; voxel mask marking quickly detects the spatial relationship between the instrument position and the restricted area through three-dimensional spatial indexing, triggering early warning signals in real time; the recommended puncture path of the nerve trunk is stored in the form of a three-dimensional coordinate point sequence, supporting real-time interpolation to generate a continuous trajectory; the coordinate sequence contains position, tangent direction and curvature information, driving the puncture needle to move smoothly along the preset trajectory; the interpolation algorithm generates a smooth transition path based on adjacent coordinate points to avoid jitter or sudden changes in the movement of the robotic arm.

[0017] Further configuration is that the S340 specifically includes the following sub-steps: S340.1. Establishing a mapping relationship between a current scanning plane and an anatomical position spatial coordinate system based on the real-time spatial positioning data of the ultrasound probe; projecting the three-dimensional navigation path grid onto a two-dimensional plane of the enhanced ultrasound image dataset using a coordinate transformation matrix; S340.2. In the cross-sectional view of the enhanced ultrasound image dataset, the following contents are superimposed and displayed: Recommended puncture path of the nerve trunk: The projected nerve trunk path is annotated with visual markers, and the path curvature is smoothly updated through a real-time interpolation algorithm; Prohibited intrusion zone of distal branches: The buffer zone projection of the distal branches is marked with a warning symbol, and a dynamic warning is triggered when the tip of the operating instrument enters the zone; Safe operating boundary: The projection range of the bounding box is marked with a boundary indicator, and the minimum distance between the operating device and the boundary is displayed in real time; S340.3. Generate a real-time visual guidance interface, wherein the real-time visual guidance interface includes the above-mentioned superimposed and annotated cross-sectional view and spatial coordinate information of the three-dimensional navigation path grid.

[0018] By adopting the above technical solution, the three-dimensional navigation grid and the real-time ultrasound image are strictly aligned through the coordinate transformation matrix to ensure the spatial consistency of the visual guidance information and the anatomical structure; the superimposed annotations of the cross-sectional view simplify the operator's operation process and avoid distraction caused by interface switching; the registration algorithm combines the spatial positioning data of the ultrasound probe to update the projection matrix in real time to compensate for coordinate offsets caused by probe movement or changes in patient position; when the tip of the instrument approaches a high-risk area or a prohibited intrusion area, the three-dimensional view triggers an enhanced warning and the cross-sectional view increases the frequency of warning symbols; the multi-level alarm mechanism significantly shortens the operator's risk response time and reduces the probability of misoperation; the warning intensity is dynamically adjusted according to the distance threshold, triggering high-frequency visual flashes and acoustic alarms at close range and displaying gentle prompts at long distances to avoid excessive interference with the operator's operation.

[0019] Further configuration is that the S340 specifically further includes the following sub-steps: S340.4. The three-dimensional coordinates of the tip of the operating instrument are collected in real time using a micro-positioning sensor, and the position of the tip of the operating instrument is marked with a three-dimensional coordinate positioning icon in the cross-sectional view and the three-dimensional view window of the anatomical position space coordinate system of the enhanced ultrasound image data set. When the tip of the operating instrument approaches the high-risk operation area or the prohibited intrusion area, a graded alarm is triggered: If the distance between the operating instrument tip and the high-risk operating area is less than a preset first threshold, an enhanced warning is displayed in the high-density nerve plexus area corresponding to the current instrument position in the three-dimensional view window of the anatomical position spatial coordinate system; If the distance between the tip of the operating instrument and the prohibited intrusion area is less than a preset second threshold, the triggering frequency of the second warning symbol in the cross-sectional view of the enhanced ultrasound image dataset increases.

[0020] By adopting the above technical solution, the micro-positioning sensor continuously collects the three-dimensional coordinates of the instrument tip, the three-dimensional view window updates its spatial position in real time, and the cross-sectional view synchronously displays the projection coordinates; multi-perspective collaborative feedback enhances the surgeon's perception of the spatial relationship of the instrument; the pulsed highlight warning in high-risk areas is combined with acoustic prompts to force the surgeon to quickly focus on key areas; this design effectively reduces the risk of misjudgment caused by visual fatigue in complex anatomical environments.

[0021] Further configuration is that the S400 specifically further includes the following sub-steps: S410: Real-time acquisition of spatial motion trajectory data of the operating instrument tip, and matching analysis with the recommended puncture path of the nerve trunk in the three-dimensional navigation path grid to generate path offset and motion stability indicators; including: The micro positioning sensor is used to obtain a three-dimensional coordinate sequence of the tip of the operating instrument at a preset sampling frequency, and the shortest distance between the tip and the center line of the recommended puncture path of the nerve trunk is calculated as the path offset; Calculating a smoothness parameter of the motion trajectory based on the acceleration and angular velocity changes of the three-dimensional coordinate sequence to generate a motion stability index; S420: generating an optimized three-dimensional navigation path grid according to the path offset and the motion stability index; including: If the path deviation exceeds a preset tolerance threshold, the local curvature of the recommended puncture path of the nerve trunk is corrected by using a Bezier curve interpolation algorithm, and the dynamic avoidance buffer radius of the terminal branch is expanded; If the motion stability index is lower than a preset safety threshold, the safety redundancy distance of the dynamic avoidance buffer zone is increased, and a temporary prohibited operation area is marked in the three-dimensional navigation path grid.

[0022] S430, updating the real-time visual guidance interface based on the optimized three-dimensional navigation path grid, and synchronously triggering a graded risk warning; including: In the cross-sectional view, a visual marker of a recommended puncture path of a nerve trunk is adjusted by a dynamic coding rule, and the display of the visual marker changes dynamically according to a real-time offset to represent a path matching degree; When the tip of the operating instrument enters the temporarily prohibited operation area, a dynamic warning symbol is superimposed on the three-dimensional view window and a warning signal is generated.

[0023] By adopting the above technical solution, the shortest distance between the tip of the operating instrument and the recommended path can be calculated in real time to generate a path offset index; the dynamic offset calculation is based on a spatial geometry algorithm to quickly solve the shortest Euclidean distance from the instrument position to the centerline of the path, and the offset trend is visualized through color gradients to provide instant feedback; based on the path offset and motion stability index, the navigation path curvature and avoidance buffer range are adjusted in real time; when a continuous offset is detected, the path optimization algorithm replans the local trajectory to bypass the obstacle area caused by tissue displacement, while expanding the safety margin to compensate for operational uncertainty.

[0024] The present invention also provides an ultrasound image-assisted anesthesia nerve precise positioning system, which includes the following modules: Ultrasonic image acquisition and enhancement module, used to acquire real-time dynamic ultrasound image sequences and perform multimodal data fusion processing to generate enhanced ultrasound image data sets; The 3D semantic segmentation module uses a cascaded convolutional neural network to segment the enhanced ultrasound image dataset and extract the morphological characteristics and spatial distribution information of the target neural tissue; The navigation path generation module constructs a three-dimensional navigation path grid based on the segmentation results, including the recommended puncture path of the nerve trunk, the prohibited intrusion area, and the safety boundary; A visual guidance module dynamically registers the 3D navigation path grid with the enhanced ultrasound image dataset, generating a real-time visual interface and overlaying instrument positioning and risk warning information; The dynamic optimization module adjusts the navigation path and risk warning strategy in real time based on operation feedback and updates the visual interface. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a main flow chart of the embodiment; Figure 2 This is a schematic diagram of the process of S100 in the embodiment; Figure 3 Schematic diagram of the process of S200 in the embodiment; Figure 4 This is a schematic diagram of the process of S300 in the embodiment; Figure 5 Schematic diagram of the process of the step of the three-dimensional navigation path grid including the spatial coordinate information of the recommended puncture path of the nerve trunk, the prohibited invasion area of ​​the peripheral branch, and the safe operation boundary in S320 of the embodiment; Figure 6 Schematic diagram of the process of S340 in the embodiment; Figure 7 Schematic diagram of the process of S400 in the embodiment. DETAILED DESCRIPTION

[0026] The present invention will be further described in detail below with reference to the accompanying drawings.

[0027] As attached Figures 1 to 7 As shown; This embodiment discloses a method for accurately locating anesthesia nerves using ultrasound imaging-assisted processing, comprising the following steps: S100, acquiring a real-time dynamic ultrasound image sequence of a target area through an ultrasound image data acquisition device, and performing multimodal data fusion processing on the real-time dynamic ultrasound image sequence to generate an enhanced ultrasound image data set; wherein the multimodal data fusion processing includes spatial resolution enhancement, noise suppression, and tissue contrast optimization; S200, performing three-dimensional semantic segmentation on the enhanced ultrasound image dataset based on a preset anesthesia nerve region segmentation model to extract morphological features and spatial distribution information of target nerve tissue; wherein the anesthesia nerve region segmentation model adopts a cascade convolutional neural network structure; S300, generating a three-dimensional navigation path grid based on the extracted morphological features and spatial distribution information, and constructing a real-time visual guidance interface; wherein the three-dimensional navigation path grid includes the spatial coordinate information of the recommended puncture path of the nerve trunk, the prohibited intrusion area of ​​the peripheral branch, and the safe operation boundary; S400, dynamically optimize navigation paths and risk warning strategies based on real-time operation feedback.

[0028] Specifically, S100 includes the following sub-steps: S110. Perform dynamic noise suppression on real-time dynamic ultrasound image sequences. By analyzing the spectral distribution characteristics of the noise in the images, an adaptive filter template based on regional sensitivity is constructed to suppress high-frequency speckle noise and low-frequency motion artifacts while preserving the continuity of nerve edges. The adaptive filter template dynamically adjusts the filter strength based on the gradient amplitude of local pixels, reducing the filter strength in dense nerve areas to protect microstructural details and increasing the filter strength in homogeneous tissue areas to eliminate noise interference. The specific implementation of dynamic noise suppression processing includes: The time-frequency characteristics of the ultrasound image sequence are analyzed by short-time Fourier transform to separate the noise-dominated frequency band from the signal-dominated frequency band. An adaptive frequency-domain attenuation function is constructed in the noise-dominated frequency band to suppress the energy density of speckle noise. Edge protection constraints are introduced in the signal-dominated frequency band to ensure the phase information integrity of the nerve boundary. The denoised image sequence is reconstructed by inverse short-time Fourier transform, and the morphological parameters of the nerve edge are verified to ensure that they meet the preset fidelity threshold.

[0029] S120. Perform multi-scale spatial resolution optimization on the noise-suppressed ultrasound image to reconstruct subtle neural structures lost due to ultrasound diffraction effects by fusing local texture features of the image with global structural information. The multi-scale optimization includes: extracting the spatial distribution pattern of neural trunks at a coarse scale to generate a neural pathway skeleton at low resolution; combining the phase information of the ultrasound echo signal at a fine scale to restore the morphological characteristics of nerve terminal branches; and fusing the coarse-scale and fine-scale reconstruction results through an iterative optimization algorithm to generate a high-fidelity neural image. Among them, the iterative optimization algorithm includes: A structural similarity metric function is defined between the coarse-scale and fine-scale reconstruction results. The structural similarity metric function is minimized by the gradient descent method, and the fusion weight coefficient is adjusted. Finally, a non-local mean filter is used to eliminate the block artifacts introduced during the iterative process. Example 1 A high-frequency ultrasound probe is used to scan the target area to obtain a real-time dynamic ultrasound image sequence. The time-frequency characteristics of the image sequence are analyzed by short-time Fourier transform to separate the noise-dominated frequency band from the signal-dominated frequency band. An adaptive frequency domain attenuation function is constructed, which is specifically in the form of: ; in, is the noise center frequency, is the frequency bandwidth, is a frequency variable; is the adaptive frequency domain attenuation function; when When the index term Inhibit The noise energy centered at hour, , that is, retaining the original signal energy and ensuring the integrity of the neural boundary.

[0030] At the coarse-scale level, a U-Net network is used to extract the spatial distribution pattern of the neural trunk and generate a low-resolution neural path skeleton with a voxel size of 0.5mm×0.5mm×1mm; at the fine-scale level, an iterative optimization algorithm is used to reconstruct the nerve terminal branches by combining the phase information of the ultrasound echo signal. The specific process is as follows: I. Define the structural similarity measurement function: ; in, represents the coarse-scale reconstruction result; represents the fine-scale reconstruction result; and Respectively and The local mean of and Respectively and The local variance of express and covariance of Represents the structural similarity index between the coarse-scale reconstruction results and the fine-scale reconstruction results; and It is a constant used to ensure numerical stability in the structural similarity calculation.

[0031] II. Adjust the fusion weight coefficient through the gradient descent method, minimize the SSIM function, and fuse the coarse / fine scale results.

[0032] III. Non-local means filtering was used with a search window of 15 × 15 and a similarity window of 5 × 5 to eliminate block artifacts and generate high-fidelity neuroimages with a voxel resolution of 0.1 mm × 0.1 mm × 0.2 mm.

[0033] S130, performs regional contrast enhancement on high-fidelity neural images, including: Divide the neural core area, vascular distribution area and background soft tissue area in high-fidelity neuroimaging, and define the grayscale dynamic range of each area; A local histogram stretching algorithm is used in the nerve core area to enhance the boundary contrast between the perineurium and myelin sheath; The overlapping artifacts between the vascular wall and the epineurium are suppressed by a nonlinear mapping function in the vascular distribution area; The nonlinear mapping function is specifically: ; in, is the input pixel grayscale value; is the output pixel grayscale value; and The typical grayscale threshold representing the blood vessel wall and the epineurium is dynamically set based on statistical prior knowledge; are the piecewise function coefficients, respectively, which are determined by minimizing the gradient conflict in the vessel-nerve overlap area; Apply global contrast compression algorithm in the background soft tissue area to reduce the visual interference of irrelevant tissues; The enhancement results of each region are spatially fused to generate an enhanced ultrasound image dataset.

[0034] The execution process of the local histogram stretching algorithm includes: A radial partition grid was established with the geometric center of the neural core as the origin. The grayscale histogram distribution was independently calculated within each grid cell of the radial partition grid, and the stretching interval was set based on the typical grayscale range of the perineurium. Finally, the grayscale transition between adjacent grid cells was smoothed using a bilinear interpolation algorithm to avoid step-like contrast changes at the boundaries. Example 2

[0035] The imaging area was divided into a neural core region with a grayscale range of 80-150 HU, a vascular distribution region with a grayscale range of 150-250 HU, and a background soft tissue region with a grayscale range of 30-80 HU. A radial partitioning grid with a sector angle of 10° was used in the neural core region to perform local histogram stretching, with the stretching interval set to [70,160] HU. A piecewise nonlinear mapping function was applied to the vascular distribution region: ; Among them, the parameters are optimized by genetic algorithm, specifically: ; Global contrast compression is used in the background area to finally generate an enhanced ultrasound image dataset.

[0036] Specifically, S200 includes the following sub-steps: S210, constructing a cascade convolutional neural network model, the cascade convolutional neural network model including a first-level coarse segmentation network, a second-level fine segmentation network, and a cross-level attention fusion module; S220, performing three-dimensional voxel processing on the enhanced ultrasound image data set to generate standardized voxel data blocks; S230, inputting the voxel data block into the cascade convolutional neural network model, performing three-dimensional semantic segmentation, generating a first-level feature map and a second-level feature map, and performing weighted fusion; S240, generating a three-dimensional neural region mask based on the weighted fusion of the first-level feature map and the second-level feature map, and optimizing the segmentation result through morphological post-processing; S250, extracting the morphological characteristics and spatial distribution information of the target neural tissue from the three-dimensional neural region mask; specifically comprising the following sub-steps: Based on the spatial connectivity of voxels in the three-dimensional neural region mask, the centerline path of the neural trunk and its curvature distribution are calculated to generate a neural direction vector field as a morphological feature. The spatial connectivity of the voxels is defined as follows: based on the voxel distribution in the three-dimensional neural region mask, the six-neighborhood or twenty-six-neighborhood rule is used to determine whether adjacent voxels belong to the same neural path; by traversing all connected voxels, a continuous path set of the neural trunk is generated.

[0037] The topological structure of candidate nerve terminal branches is extracted through a three-dimensional skeletonization algorithm to generate three-dimensional skeletonized branches as spatial distribution information, and the branch length, bifurcation angle and diameter change rate are quantified. The implementation of the three-dimensional skeletonization algorithm includes: iteratively removing the outer voxels of the candidate nerve terminal branches based on a morphological thinning algorithm until only the center line of a single voxel width is retained; and extracting the three-dimensional path of the center line through topological connection rules to generate a three-dimensional skeletonized branch.

[0038] Integrating the spatial coordinate system defined in the enhanced ultrasound image dataset, the nerve trajectory vector field and three-dimensional skeletonized branches were mapped to corresponding anatomical locations. A heat map of the neural tissue density distribution was constructed, and the geometric center coordinates of high-density neural plexus regions were annotated. High-density neural plexus regions were defined by voxel clusters whose probability density values ​​exceeded a preset threshold in the heat map. The spatial coordinate system was defined as follows: Based on the voxel spatial coordinate information recorded in the three-dimensional voxel data blocks of the enhanced ultrasound image dataset, a Cartesian coordinate system with the initial scanning position of the ultrasound probe as the origin was established. The coordinates of each voxel in the coordinate system were aligned with the actual physical spatial position using an interpolation algorithm.

[0039] Specifically, S220 includes: The axial slice sequence is divided along the ultrasound probe scanning direction, and an isotropic voxel grid is constructed based on spatial continuity constraints between adjacent slices. The grayscale values ​​within each voxel grid are optimized using bicubic interpolation to eliminate inter-slice resolution differences caused by probe motion. The optimized voxel grid is integrated through three-dimensional spatial resampling to generate standardized voxel data blocks, where each voxel contains spatial coordinate information and corresponding grayscale intensity value. S230 specifically includes: The first-level coarse segmentation network extracts global contextual features of each axial slice and generates a first-level feature map containing the neural trunk area. The second-level fine segmentation network refines the first-level feature map layer by layer, uses dilated convolution to expand the receptive field to cover the topological structure of the neural branches, and retains low-level detail information through residual connection to output the second-level feature map. In the cross-level attention fusion module, the channel correlation matrix of the first-level feature map and the second-level feature map is calculated to generate a spatial attention weight map. Based on the spatial attention weight map, the first-level feature map and the second-level feature map are weightedly fused to highlight the semantic information of the neural core area. Morphological post-processing in S240 includes: The weighted fusion of the first-level feature map and the second-level feature map is thresholded to generate a 3D neural region mask. Isolated noise points in the 3D neural region mask are removed, retaining continuous neural pathways. After this, the neural path boundary is smoothed to eliminate jagged edges caused by ultrasound artifacts. Ultrafine branches in the 3D neural region mask are then filtered based on a preset nerve diameter threshold, retaining candidate nerve terminal branches and eliminating false neural structures caused by artifact interference. Candidate nerve terminal branches are defined as: neural branches in the 3D neural region mask with a diameter less than a preset threshold are filtered, retaining their connectivity structure as candidate nerve endings. Example 3

[0040] Construct a cascaded network structure: the first-level coarse segmentation network uses a 3D ResNet-18 architecture, with an input voxel data block size of 128×128×64, and outputs a neural trunk region feature map; the second-level fine segmentation network uses a 3D U-Net architecture with a dilated convolutional layer, which outputs a fine-grained neural branch feature map; the cross-level attention fusion module calculates the channel correlation matrix, generates a spatial attention weight map, and weighted fusion feature maps; The fused feature map is threshold segmented with a threshold of 0.65 to generate a 3D neural region mask. 3D connected domain analysis is used to remove isolated noise points; a 3D Gaussian filter is used with the configuration The boundaries were smoothed by voxels, and candidate terminal branches were screened based on the nerve diameter threshold, with the trunk ≥ 1 mm and the terminal < 0.5 mm.

[0041] Specifically, S300 includes the following sub-steps: S310: Generate a navigation baseline for the nerve trunk and an avoidance path for the distal branches using a path optimization algorithm based on the curvature distribution of the nerve trajectory vector field and the topological structure of the three-dimensional skeletonized branches. The path optimization algorithm includes: Equally spaced sampling points are set along the centerline path of the nerve trunk, and the curvature radius of each sampling point is calculated. If the curvature radius is less than a preset safety threshold, the path curvature is smoothed using a Bezier curve interpolation algorithm to avoid sudden changes in the navigation path. The specific application of Bezier curve interpolation algorithm is as follows: When the radius of curvature of the nerve trunk centerline path is detected to be less than a preset safety threshold, the algorithm regenerates a smooth navigation path by inserting Bezier curve control points. For example, using quadratic or cubic Bezier curves, the algorithm dynamically adjusts the path geometry to eliminate discontinuities or dangerous bends in the navigation path.

[0042] Based on the bifurcation angle and diameter change rate of the three-dimensional skeletonized branches, a dynamic avoidance buffer zone is constructed around the terminal branches. The radius of the buffer zone is positively correlated with the diameter of the terminal branches. S320: Mapping the navigation baseline and the avoidance path to the spatial coordinate system of the anatomical position to generate a three-dimensional navigation path grid; wherein the three-dimensional navigation path grid includes spatial coordinate information of the recommended puncture path of the nerve trunk, the prohibited intrusion area of ​​the peripheral branch, and the safe operation boundary; S330, based on the geometric center coordinates of the high-density neural plexus region in the density distribution heat map, marking the high-risk operation area in the three-dimensional navigation path grid, and generating a risk level label through probability density gradient calculation; wherein the high-risk operation area is defined as a voxel cluster whose probability density value exceeds a preset threshold in the spatial coordinate system; the risk level label includes low risk, medium risk, and high risk; S340. Dynamically spatially register the three-dimensional navigation path grid with the enhanced ultrasound image data set to generate a real-time visual guidance interface.

[0043] Specifically, in S320, the three-dimensional navigation path grid includes the spatial coordinate information of the recommended puncture path of the nerve trunk, the prohibited invasion area of ​​the peripheral branch, and the safe operation boundary, specifically: The recommended puncture path of the nerve trunk is generated by the centerline path of the nerve trend vector field after curvature smoothing, and is represented by a sequence of three-dimensional coordinate points in the spatial coordinate system; The prohibited intrusion area of ​​the terminal branch is defined as a cylindrical area with the center line of the terminal branch as the axis and the buffer zone radius as the boundary according to the radius of the dynamic avoidance buffer zone and the spatial position of the terminal branch, and is marked by a voxel mask in the spatial coordinate system; The safe operation boundary is generated based on the minimum safe distance threshold between the nerve trunk and the terminal branches, and is marked in the form of a bounding box in the spatial coordinate system. The size of the bounding box is determined by the average diameter of the nerve trunk and the density distribution heat map of the high-density nerve plexus area.

[0044] Specifically, S340 includes the following sub-steps: S340.1. Based on the real-time spatial positioning data of the ultrasound probe, establish a mapping relationship between the current scanning plane and the anatomical position spatial coordinate system; project the three-dimensional navigation path grid onto the two-dimensional plane of the enhanced ultrasound image dataset using a coordinate transformation matrix; S340.2. In the cross-sectional view of the enhanced ultrasound image data set, the following are superimposed and displayed: Recommended puncture path of the nerve trunk: The projected nerve trunk path is annotated with visual markers, and the path curvature is smoothly updated using a real-time interpolation algorithm; specifically, the visual markers are marked as blue dotted lines; Prohibited intrusion zone of distal branches: A cylindrical buffer zone projection of distal branches is marked with a warning symbol, and a dynamic warning is triggered when the tip of the operating instrument enters the zone; specifically, the warning symbol is marked as a red semi-transparent area mark; Safe operating boundary: The projection range of the bounding box is marked with a boundary indicator, and the minimum distance between the operating device and the boundary is displayed in real time; specifically, the boundary indicator is marked with a green dotted box.

[0045] S340.3. Generate a real-time visual guidance interface, the real-time visual guidance interface including the above-mentioned superimposed annotated cross-sectional view and the spatial coordinate information of the three-dimensional navigation path grid; Specifically, S340 further includes the following sub-steps: S340.4. The three-dimensional coordinates of the instrument tip are collected in real time using a micro-positioning sensor, and the position of the instrument tip is marked with a three-dimensional coordinate positioning icon in the cross-sectional view of the enhanced ultrasound image data set and the three-dimensional view window of the anatomical position space coordinate system. When the instrument tip approaches a high-risk operation area or a prohibited intrusion area, a graded alarm is triggered: If the distance between the tip of the operating instrument and the high-risk operating area is less than a preset first threshold, an enhanced warning is displayed in the high-density nerve plexus area corresponding to the current instrument position in the three-dimensional view window of the anatomical position space coordinate system; If the distance between the operating instrument tip and the prohibited intrusion area is less than a preset second threshold, the triggering frequency of the second warning symbol increases in the cross-sectional view of the enhanced ultrasound image data set, accompanied by an acoustic prompt.

[0046] Specifically, if the distance between the tip of the instrument and the high-risk operation area or the prohibited intrusion area is calculated as: ; in, The distance between the tip of the operating instrument and the high-risk operating area or the prohibited intrusion area; is the coordinate of the tip of the operating instrument, The center coordinates of the high-risk operation area or the prohibited intrusion area. Example 4

[0047] In the path optimization algorithm, a cubic Bezier curve is inserted into the sampling points with a curvature radius of less than 3mm, and the control point spacing is 2mm to ensure the curvature continuity; in the dynamic avoidance buffer zone, the radius of the terminal branch buffer zone is ,in is the branch radius, constructing a cylindrical restricted area; The safe operation boundary is generated by calculating the Hausdorff distance between the neural trunk and the terminal branches to generate a bounding box; In the real-time visualization guide interface, the cross-sectional view is overlaid as follows: Recommended puncture path of the nerve trunk: blue dotted line, with a line width of 2px and a transparency of 0.7; No trespassing zone: red translucent cylinder with a transparency of 0.3; Safety border: green dashed box with a line width of 1px.

[0048] When the tip of the instrument is less than 1 mm away from the high-risk area, a pulse warning box is triggered with a frequency of 2 Hz and RGB set to (255, 0, 0).

[0049] Specifically, S400 further includes the following sub-steps: S410: Real-time acquisition of spatial motion trajectory data of the instrument tip, and matching analysis with the recommended puncture path of the nerve trunk in the three-dimensional navigation path grid to generate path offset and motion stability indicators; including: The three-dimensional coordinate sequence of the tip of the operating instrument is acquired by a micro positioning sensor at a preset sampling frequency, and the shortest distance between the tip and the center line of the recommended puncture path of the nerve trunk is calculated as the path offset; Based on the changes in acceleration and angular velocity of the three-dimensional coordinate sequence, the smoothness parameter of the motion trajectory is calculated to generate a motion stability index; the smoothness parameter is determined by Fourier transform analysis of the high-frequency noise energy ratio in the trajectory spectrum; S420, generating an optimized three-dimensional navigation path grid according to the path offset and the motion stability index; including: If the path deviation exceeds the preset tolerance threshold, the local curvature of the recommended puncture path of the nerve trunk is corrected through the Bezier curve interpolation algorithm, and the dynamic avoidance buffer radius of the terminal branch is expanded; If the motion stability index is lower than the preset safety threshold, the safety redundancy distance of the dynamic avoidance buffer zone is increased, and a temporary prohibited operation area is marked in the three-dimensional navigation path grid.

[0050] S430: Update the real-time visual guidance interface based on the optimized three-dimensional navigation path grid and simultaneously trigger a graded risk warning; including: In the cross-sectional view, the visual markers of the recommended puncture paths of the nerve trunk are adjusted through dynamic coding rules. The display of the visual markers changes dynamically according to the real-time offset to represent the path matching degree. Specifically, the optimized recommended puncture paths of the nerve trunk are marked with dynamic color coding, and the path color gradually changes from green to red according to the real-time offset. When the tip of the operating instrument enters a temporarily prohibited operation area, a dynamic warning symbol is superimposed in the three-dimensional view window and a warning signal is generated; specifically, a pulse-type highlighted warning box is superimposed and displayed in the three-dimensional view window, and the tactile feedback device is activated to generate a vibration warning signal. Example 5

[0051] If the path deviation is greater than 2mm, expand the avoidance buffer radius to , and insert additional Bezier control points with a spacing shortened to 1 mm; If the motion stability index is greater than 0.3, a temporary restricted area is marked and the safety distance is extended to 3mm. Example 6

[0052] Clinical tests have shown that the nerve positioning accuracy of this method reaches 0.8±0.2mm, and the warning response time for entering a restricted area is less than 50ms, which is significantly better than the traditional ultrasound guidance method with an accuracy of 1.5±0.5mm.

[0053] This embodiment also discloses an ultrasound image-assisted anesthetic nerve precise positioning system, which includes the following modules: Ultrasonic image acquisition and enhancement module, used to acquire real-time dynamic ultrasound image sequences and perform multimodal data fusion processing to generate enhanced ultrasound image data sets; The 3D semantic segmentation module uses a cascaded convolutional neural network to segment the enhanced ultrasound image dataset and extract the morphological characteristics and spatial distribution information of the target neural tissue; The navigation path generation module constructs a three-dimensional navigation path grid based on the segmentation results, including the recommended puncture path of the nerve trunk, the prohibited intrusion area, and the safety boundary; A visual guidance module dynamically registers the 3D navigation path grid with the enhanced ultrasound image dataset, generating a real-time visual interface and overlaying instrument positioning and risk warning information; The dynamic optimization module adjusts the navigation path and risk warning strategy in real time based on operation feedback and updates the visual interface.

[0054] This specific embodiment is merely an explanation of the present invention and is not intended to limit the present invention. After reading this specification, those skilled in the art may make non-creative modifications to this embodiment as needed. However, as long as such modifications are within the scope of the claims of the present invention, they are protected by patent law.

Claims

1. A method for accurately locating anesthesia nerves using ultrasound imaging assistance, characterized in that: The following steps are included: S100, acquiring a real-time dynamic ultrasound image sequence of a target area through an ultrasound image data acquisition device, and performing multimodal data fusion processing on the real-time dynamic ultrasound image sequence to generate an enhanced ultrasound image dataset; wherein the multimodal data fusion processing includes spatial resolution enhancement, noise suppression, and tissue contrast optimization; S200, performing three-dimensional semantic segmentation on the enhanced ultrasound image dataset based on a preset anesthesia nerve region segmentation model to extract morphological features and spatial distribution information of target nerve tissue; wherein the anesthesia nerve region segmentation model adopts a cascade convolutional neural network structure; S300, generating a three-dimensional navigation path grid based on the extracted morphological features and spatial distribution information, and constructing a real-time visual guidance interface; wherein the three-dimensional navigation path grid includes the spatial coordinate information of the recommended puncture path of the nerve trunk, the prohibited intrusion area of ​​the peripheral branch, and the safe operation boundary; S400, dynamically optimize navigation paths and risk warning strategies based on real-time operation feedback.

2. The method for accurately locating anesthesia nerves using ultrasound imaging-assisted processing according to claim 1, characterized in that: The S100 specifically includes the following sub-steps: S110, performing dynamic noise suppression processing on the real-time dynamic ultrasound image sequence, constructing an adaptive filtering template based on regional sensitivity by analyzing the spectral distribution characteristics of the noise in the image; wherein the adaptive filtering template dynamically adjusts the filtering strength according to the gradient amplitude of the local pixel, reducing the filtering strength in nerve-dense areas to protect microstructural details, and increasing the filtering strength in homogeneous tissue areas to eliminate noise interference; S120, performing multi-scale spatial resolution optimization on the noise-suppressed ultrasound image to fuse local texture features and global structural information of the image; wherein the multi-scale optimization includes: extracting the spatial distribution pattern of the nerve trunk at a coarse scale level to generate a neural pathway skeleton at a low resolution; combining the phase information of the ultrasound echo signal at a fine scale level to restore the morphological characteristics of the nerve terminal branches; and fusing the coarse-scale and fine-scale reconstruction results through an iterative optimization algorithm to generate a high-fidelity neural image; S130, performing region-by-region contrast enhancement processing on the high-fidelity neural image, including: dividing the high-fidelity neural image into a neural core area, a vascular distribution area, and a background soft tissue area, and defining the grayscale dynamic range of each area; A local histogram stretching algorithm is used in the nerve core area to enhance the boundary contrast between the perineurium and myelin sheath; The overlapping artifacts between the vascular wall and the epineurium are suppressed by a nonlinear mapping function in the vascular distribution area; Apply global contrast compression algorithm in the background soft tissue area to reduce the visual interference of irrelevant tissues; The enhancement results of each region are spatially fused to generate the enhanced ultrasound image dataset.

3. The method for accurately locating anesthesia nerves using ultrasound imaging-assisted processing according to claim 2, characterized in that: The S200 specifically includes the following sub-steps: S210, constructing a cascade convolutional neural network model, wherein the cascade convolutional neural network model includes a first-level coarse segmentation network, a second-level fine segmentation network, and a cross-level attention fusion module; S220, performing three-dimensional voxel processing on the enhanced ultrasound image data set to generate standardized voxel data blocks; S230, inputting the voxel data block into the cascade convolutional neural network model, performing three-dimensional semantic segmentation, generating a first-level feature map and a second-level feature map, and performing weighted fusion; S240, generating a three-dimensional neural region mask based on the weighted fusion of the first-level feature map and the second-level feature map, and optimizing the segmentation result through morphological post-processing; S250, extracting morphological features and spatial distribution information of target neural tissue from the three-dimensional neural region mask; It includes the following sub-steps: Based on the spatial connectivity of voxels in the three-dimensional neural region mask, the centerline path of the neural trunk and its curvature distribution are calculated to generate a neural direction vector field as a morphological feature; Extracting the topological structure of the candidate nerve terminal branches using a three-dimensional skeletonization algorithm to generate three-dimensional skeletonized branches as spatial distribution information, and quantifying branch lengths, bifurcation angles, and diameter change rates; Combined with the spatial coordinate system defined in the enhanced ultrasound image dataset, the nerve direction vector field and the three-dimensional skeletonized branches are mapped to the corresponding anatomical positions, a density distribution heat map of the neural tissue is constructed, and the geometric center coordinates of the high-density neural plexus area are marked; wherein the high-density neural plexus area is defined by a voxel cluster whose probability density value exceeds a preset threshold in the heat map.

4. The method for accurately locating anesthesia nerves using ultrasound imaging-assisted processing according to claim 3, characterized in that: The S220 specifically includes: The axial slice sequence is divided along the ultrasound probe scanning direction, and an isotropic voxel grid is constructed based on spatial continuity constraints between adjacent slices. The grayscale values ​​within each voxel grid are optimized using bicubic interpolation to eliminate inter-slice resolution differences caused by probe motion. The optimized voxel grid is integrated through three-dimensional spatial resampling to generate standardized voxel data blocks, where each voxel contains spatial coordinate information and corresponding grayscale intensity value. The S230 specifically includes: The first-level coarse segmentation network extracts global contextual features of each axial slice to generate a first-level feature map containing the neural trunk area; the second-level fine segmentation network refines the first-level feature map layer by layer, uses dilated convolution to expand the receptive field to cover the topological structure of the neural branches, retains low-level detail information through residual connection, and outputs a second-level feature map; in the cross-level attention fusion module, the channel correlation matrix of the first-level feature map and the second-level feature map is calculated to generate a spatial attention weight map; based on the spatial attention weight map, the first-level feature map and the second-level feature map are weightedly fused to highlight the semantic information of the neural core area; The morphological post-processing in S240 includes: The first-level feature map and the second-level feature map after weighted fusion are threshold segmented to generate a three-dimensional neural region mask; isolated noise points in the three-dimensional neural region mask are removed, and continuous neural paths are retained; then, ultra-fine branches of the three-dimensional neural region mask are screened based on a preset nerve diameter threshold, candidate nerve terminal branches are retained, and false neural structures formed by artifact interference are eliminated.

5. The method for accurately locating anesthesia nerves using ultrasound imaging-assisted processing according to claim 3, characterized in that: The S300 specifically includes the following sub-steps: S310: Generate a navigation baseline for the nerve trunk and an avoidance path for the distal branches using a path optimization algorithm based on the curvature distribution of the nerve trajectory vector field and the topological structure of the three-dimensional skeletonized branches; wherein the path optimization algorithm includes: Equally spaced sampling points are set along the centerline path of the nerve trunk, and the curvature radius of each sampling point is calculated. If the curvature radius is less than a preset safety threshold, the path curvature is smoothed using a Bezier curve interpolation algorithm to avoid sudden changes in the navigation path. Constructing a dynamic avoidance buffer zone around the distal branch according to the bifurcation angle and diameter change rate of the three-dimensional skeletonized branch, wherein the radius of the buffer zone is positively correlated with the diameter of the distal branch; S320: Mapping the navigation baseline and the avoidance path to the spatial coordinate system of the anatomical position to generate a three-dimensional navigation path grid; wherein the three-dimensional navigation path grid includes spatial coordinate information of the recommended puncture path of the nerve trunk, the prohibited intrusion area of ​​the peripheral branch, and the safe operation boundary; S330: Based on the geometric center coordinates of the high-density nerve plexus region in the density distribution heat map, marking the high-risk operation area in the three-dimensional navigation path grid, and generating a risk level label through probability density gradient calculation; wherein the high-risk operation area is defined as a voxel cluster whose probability density value in the spatial coordinate system exceeds a preset threshold; S340: Dynamically spatially register the three-dimensional navigation path grid with the enhanced ultrasound image data set to generate a real-time visual guidance interface.

6. The method for accurately locating anesthesia nerves using ultrasound imaging-assisted processing according to claim 5, characterized in that: In S320, the three-dimensional navigation path grid includes the spatial coordinate information of the recommended puncture path of the nerve trunk, the prohibited invasion area of ​​the peripheral branch, and the safe operation boundary, specifically: The recommended puncture path of the nerve trunk is generated by curvature smoothing the center line path of the nerve trend vector field and is represented by a sequence of three-dimensional coordinate points in the spatial coordinate system; The prohibited intrusion area of ​​the terminal branch is defined as a cylindrical area with the center line of the terminal branch as the axis and the radius of the buffer zone as the boundary according to the radius of the dynamic avoidance buffer zone and the spatial position of the terminal branch, and is marked by a voxel mask in the spatial coordinate system; The safe operation boundary is generated based on the minimum safe distance threshold between the nerve trunk and the terminal branches, and is marked in the form of a bounding box in the spatial coordinate system. The size of the bounding box is determined by the average diameter of the nerve trunk and the density distribution heat map of the high-density nerve plexus area.

7. The method for accurately locating anesthesia nerves using ultrasound imaging-assisted processing according to claim 5, characterized in that: The S340 specifically includes the following sub-steps: S340.

1. Establishing a mapping relationship between a current scanning plane and an anatomical position spatial coordinate system based on the real-time spatial positioning data of the ultrasound probe; projecting the three-dimensional navigation path grid onto a two-dimensional plane of the enhanced ultrasound image dataset using a coordinate transformation matrix; S340.

2. In the cross-sectional view of the enhanced ultrasound image dataset, the following contents are superimposed and displayed: Recommended puncture path of the nerve trunk: The projected nerve trunk path is annotated with visual markers, and the path curvature is smoothly updated through a real-time interpolation algorithm; Prohibited intrusion zone of distal branches: The buffer zone projection of the distal branches is marked with a warning symbol, and a dynamic warning is triggered when the tip of the operating instrument enters the zone; Safe operating boundary: The projection range of the bounding box is marked with a boundary indicator, and the minimum distance between the operating device and the boundary is displayed in real time; S340.

3. Generate a real-time visual guidance interface, wherein the real-time visual guidance interface includes the above-mentioned superimposed marked cross-sectional view and the spatial coordinate information of the three-dimensional navigation path grid.

8. The method for accurately locating anesthesia nerves using ultrasound imaging-assisted processing according to claim 7, characterized in that: The S340 specifically further includes the following sub-steps: S340.

4. The three-dimensional coordinates of the tip of the operating instrument are collected in real time using a micro-positioning sensor, and the position of the tip of the operating instrument is marked with a three-dimensional coordinate positioning icon in the cross-sectional view and the three-dimensional view window of the anatomical position space coordinate system of the enhanced ultrasound image data set. When the tip of the operating instrument approaches the high-risk operation area or the prohibited intrusion area, a graded alarm is triggered: If the distance between the operating instrument tip and the high-risk operating area is less than a preset first threshold, an enhanced warning is displayed in the high-density nerve plexus area corresponding to the current instrument position in the three-dimensional view window of the anatomical position spatial coordinate system; If the distance between the tip of the operating instrument and the prohibited intrusion area is less than a preset second threshold, the triggering frequency of the second warning symbol in the cross-sectional view of the enhanced ultrasound image dataset increases.

9. The method for accurately locating anesthesia nerves using ultrasound imaging-assisted processing according to claim 1, characterized in that: The S400 specifically further includes the following sub-steps: S410: Real-time acquisition of spatial motion trajectory data of the operating instrument tip, and matching analysis with the recommended puncture path of the nerve trunk in the three-dimensional navigation path grid to generate path offset and motion stability indicators; including: The micro positioning sensor is used to obtain a three-dimensional coordinate sequence of the tip of the operating instrument at a preset sampling frequency, and the shortest distance between the tip and the center line of the recommended puncture path of the nerve trunk is calculated as the path offset; Calculating a smoothness parameter of the motion trajectory based on the acceleration and angular velocity changes of the three-dimensional coordinate sequence to generate a motion stability index; S420: generating an optimized three-dimensional navigation path grid according to the path offset and the motion stability index; including: If the path deviation exceeds a preset tolerance threshold, the local curvature of the recommended puncture path of the nerve trunk is corrected by using a Bezier curve interpolation algorithm, and the dynamic avoidance buffer radius of the terminal branch is expanded; If the motion stability index is lower than a preset safety threshold, the safety redundancy distance of the dynamic avoidance buffer zone is increased, and a temporary prohibited operation area is marked in the three-dimensional navigation path grid; S430, updating the real-time visual guidance interface based on the optimized three-dimensional navigation path grid, and synchronously triggering a graded risk warning; including: In the cross-sectional view, a visual marker of a recommended puncture path of a nerve trunk is adjusted by a dynamic coding rule, and the display of the visual marker changes dynamically according to a real-time offset to represent a path matching degree; When the tip of the operating instrument enters the temporarily prohibited operation area, a dynamic warning symbol is superimposed on the three-dimensional view window and a warning signal is generated.

10. An ultrasound image-assisted anesthetic nerve precise positioning system, applied to the ultrasound image-assisted anesthetic nerve precise positioning method according to any one of claims 1 to 9, characterized in that: Includes the following modules: Ultrasonic image acquisition and enhancement module, used to acquire real-time dynamic ultrasound image sequences and perform multimodal data fusion processing to generate enhanced ultrasound image data sets; The 3D semantic segmentation module uses a cascaded convolutional neural network to segment the enhanced ultrasound image dataset and extract the morphological characteristics and spatial distribution information of the target neural tissue; The navigation path generation module constructs a three-dimensional navigation path grid based on the segmentation results, including the recommended puncture path of the nerve trunk, the prohibited intrusion area, and the safety boundary; A visual guidance module dynamically registers the 3D navigation path grid with the enhanced ultrasound image dataset, generating a real-time visual interface and overlaying instrument positioning and risk warning information; The dynamic optimization module adjusts the navigation path and risk warning strategy in real time based on operation feedback and updates the visual interface.

Images